By Bubacarr Bah
The African Institute for Mathematical Sciences (AIMS) South Africa hosted the 11th Gene Golub SIAM Summer School (G2S3) from July 19-30, 2021. The event—which was originally scheduled for 2020 but got postponed due to the ongoing COVID-19 pandemic—took place mostly virtually, which allowed for the inclusion of additional attendees from African institutions. The theme of the 2021 school was “Theory and Practice of Deep Learning,” with a focus on the theory, implementation, and application of neural networks. Course material exposed students to math-based concepts from functional and harmonic analysis and optimization theory, which they used to explore the mathematical underpinnings of deep learning. Participants also experimented with multiple deep learning applications, such as computer vision and forecasting, and listened to additional lectures by industry practitioners who work on real-life problems.
Because the 2021 school was the first iteration of G2S3 to take place in Africa, it garnered a lot of interest. A total of 46 attendees represented 26 nationalities from 17 different countries around the world. About 50 percent of participants were African, and nearly all of these students were located in Africa; this was thus the first school of its kind to have such significant participation from Africa itself. Most participants were graduate students in either Ph.D. or M.Sc. programs, and roughly 31 percent were female. Figures 1 and 2 provide a more detailed breakdown of attendance.
The school was held in a hybrid format due to COVID-19. Although most attendees participated virtually, several in-person cohorts did work together at the AIMS centres in Rwanda and Cameroon. The lectures and tutorials took place on Zoom and the social sessions transpired on Wonder. Daniel Nickelsen (AIMS South Africa) served as the moderator.
Barry Green, the director of AIMS South Africa, delivered the opening remarks that marked the beginning of G2S3 activities. As with any virtual event, time zone differences added a logistical complication. All lectures were scheduled for the evening hours in Africa and Europe, which corresponded to the morning and early afternoon in North America. Lecturers included Bubacarr Bah (AIMS South Africa), Coralia Cartis (University of Oxford), Gitta Kutyniok (Ludwig Maximilian University of Munich), Kasso Okoudjou (Tufts University), and Jared Tanner (University of Oxford).
The Organizing Committee—which consisted of Bah, Cartis, and Okoudjou—arranged tutorials in the afternoon for cohorts in Africa and Europe that were then repeated for the North American cohort at a later time. A social event for all cohorts in the form of a coffee/tea break took place between the two main lectures every day. Students, tutors, and lecturers were invited to bring their coffee, tea, or beverage of choice and spend some time interacting with each other on Wonder. Despite modest participation (likely due to fatigue from excessive screen time), the virtual gatherings nonetheless generated a sense of international interaction between participants. In contrast, the high-quality lectures, tutorials, demonstrations, practicals, and exercises were extremely interactive. The students greatly appreciated and benefited from these learning opportunities and were thankful for the chance to connect with fellow peers around the world.
At the school’s conclusion, we solicited feedback from attendees to gauge their levels of satisfaction with both the program and implementation. The response was overwhelmingly positive. “It was very useful to have such a wide range of depths of neural networks, which definitely appealed to the diverse backgrounds of the students,” one attendee said. Another praised the format of the tutorials, adding that there was “enough [information] to represent the breadth of the field, but not so much that students would get lost in the material.” A third student reflected on the structure of the school as a whole. “The thought put into the careful planning—accommodating various time zones and attaching practicals to each lecture—was an incredible amount of work,” the participant said. “Yet it appeared seamless.”
In summation, the 2021 school—the first G2S3 installment to be hosted by an African institution—commenced successfully in a hybrid format, with some in-person participation at the AIMS centres in Rwanda and Cameroon. It exposed participants to the state of the art of the mathematics of deep learning.
Acknowledgments: We would like to extend a special thanks to our sponsors: SIAM, the BMBF (German Federal Ministry of Education and Research), and the Alexander von Humboldt Foundation. We are also grateful for the numerous tutors who interacted with participants throughout the duration of the school.
The 2022 Gene Golub SIAM Summer School on Financial Analytics: Networks, Learning, and High Performance Computing will take place at the Gran Sasso Science Institute in L’Aquila, Italy, from August 1-12, 2022. The application portal to participate is now available online; all applications are due by April 15th.
Interested in organizing a future school? Letters of intent that propose topics and organizers for the 2024 iteration of G2S3 are due by January 31, 2023. Visit the G2S3 website to learn more.
Bubacarr Bah is a senior researcher—designated as the German Research Chair of Mathematics with a specialization in data science—at the African Institute for Mathematical Sciences South Africa and a senior lecturer (assistant professor) in Stellenbosch University’s Division of Applied Mathematics.